Penn AI tool uses motor imitation to accelerate autism screening

Researchers at the University of Pennsylvania have developed an AI-driven tool designed to speed up the autism screening process for children. By analyzing motor imitation, the system aims to reduce the significant wait times families often face when seeking diagnostic support.

Developed by Penn Engineering in collaboration with the Kennedy Krieger Institute, the tool is called CAMI, or Computerized Assessment of Motor Imitation. The system, specifically the CAMI-2DNet iteration, analyzes video footage of a child attempting to mimic specific movements. It then generates an objective score based on the quality of that imitation to help flag potential autism. In a 2025 clinical study involving 183 children, the tool demonstrated an 80% true positive rate in distinguishing autistic children from their neurotypical peers.

The role of motor imitation

Traditional autism evaluations are typically lengthy processes. They rely on a combination of parent questionnaires, developmental histories, and direct behavioral observations by clinicians. Because these manual reviews are time-intensive, many children face long delays before receiving a formal diagnosis.

Difficulty with motor imitation is a recognized behavioral marker for autism. The Penn team leveraged this marker to create an automated method that removes much of the subjectivity found in manual observations. To ensure the tool remains reliable across different settings, the AI is engineered to be robust against environmental variables. According to Ph.D. candidate Kaleab Kinfu, the system is trained to disregard "environmental noise"—such as lighting, camera angles, and the physical size of the child—to focus exclusively on the quality of the movement replication.

Impact on clinical workflows

By providing a scalable and objective screening mechanism, CAMI could significantly shorten the path to diagnosis in both school and clinic settings. The tool is not intended to replace the expertise of a physician; rather, it serves as a complement to professional evaluations. By acting as a high-efficiency filter, it allows clinicians to identify and prioritize children who require full diagnostic support more quickly than current methods allow.

Future outlook

While the initial study showed promising results among children between the ages of seven and thirteen, the tool represents a shift toward more accessible, data-driven screening. The next steps for the technology will likely involve further validation across broader age groups and integration into existing clinical pipelines. For now, the focus remains on providing a reliable, objective tool that can bridge the gap between initial suspicion and formal medical intervention.